Journey result: Journey E2E quality audit — Wave 4

Walked: 2026-06-05 as an adversarial source-and-runtime review of the then-current 52-journey suite. The original pass used six independent, read-only audits and worked its confirmed fixes through the live web/BFF stack.
Reconciled: 2026-07-18 against the current 64-journey registry, the Wave 5 and Wave 6 successor audits, and the surviving operator, PWA, provenance, and Veritas/Nisaba specs. This was a source reconciliation, not a fresh full-suite runtime walk.
Verdict: partial — Wave 4 remains useful evidence that one broken spec, two false-positive paths, grade drift, and cross-link debt were found and repaired. It is not the current coverage inventory: later waves corrected several of its premises, added real-HTTP coverage, and the registry has grown from 52 to 64 journeys.
Successors: Wave 5 pressure-tested auditor assumptions; Wave 6 superseded Wave 5 as the live work list and shipped the next repair set.

Result at a glance#

| Wave 4 lane | Dated finding | Current source disposition | Authority limit | | ---------------------- | ------------------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------- | | Editorial decision | The spec used a non-admin session and expected retired “awaiting audit replay” copy, so the real admin POST failed | Closed: the spec reissues admin inbox traffic with an admin bearer and asserts data-decision-audit-id plus Decision recorded · audit | Source presence; this reconciliation did not rerun the app spec | | PWA cache purge | currentCachePresent | | staleCacheAbsent could pass without proving eviction | Closed: the browser suite drives real service-worker cache removal, and the policy unit suite fires the activate listener against stale/current/foreign caches | Deterministic browser/unit harness, not every installed-browser race | | Provenance inspector | An if (inspectorVisible) block silently skipped the hash-chain core when no ledger existed | Closed and deepened: the current suite seeds test-only clean and broken ledgers and separately asserts unavailable and empty states | Harness ledger, not a production persistence audit | | Veritas source lineage | Later Wave 6 found a test aimed at an unmounted overlay | Closed after Wave 4: the current spec opens the live Source Directory profile and follows its Nisaba lineage action | Proves the mounted affordance, not the whole evidence graph | | Generic headers | Roughly 30 specs used non-specific “Backs the journey” prose | Closed in the current tree: no E2E spec retains the old generic header phrases | Textual traceability, not behavioral depth by itself | | Grade snapshot | Several Wave 4 deep/partial labels disagreed | Superseded: current coverage.md is authoritative; later coverage re-earned deep for install/update, first-time, Tara, and multi-device while offline and both Veritas journeys remain partial | A registry grade summarizes named evidence; it is not a fresh suite result |

Evidence map#

Wave 4 is one dated link in an evidence chain. Solid edges are preserved source proof; the dashed edge is the runtime claim this documentation audit does not make.

flowchart LR A[Wave 4 adversarial findings 52 journeys] --> B[Same-day repairs broken and false-green paths] B --> C[Wave 5 live premise checks guards, claims, failure modes] C --> D[Wave 6 route and test repairs real-HTTP specs] D --> E[Current coverage registry 64 journeys] E -. requires a fresh run .-> F[Current full-suite runtime verdict]

Proven observations#

The decisive defects were repaired, not merely relabelled#

operator-editorial-review-approval.spec.ts now authenticates the admin-gated decision request, observes the audit-bearing confirmation, retains the audit identifier, and checks the decision after reload. The obsolete “awaiting audit replay” expectation is gone.
pwa-lifecycle-deepening.spec.ts distinguishes the apply handshake from real activation eviction. It registers the actual worker in Chromium, creates current and stale caches, activates the update, and requires the stale cache to disappear. pwa-service-worker-policy.test.ts independently exercises the worker's activate handler and preserves current and foreign caches.
citation-drift-resolution.spec.ts no longer hides its core behind inspectorVisible. A test-only ledger fixture drives exact clean-chain and broken-chain verdicts; separate cases make unavailable, empty, and unknown bundle behavior visible rather than green-by-omission.
The current E2E tree has no “Backs the journey above” or “Backs the journeys above” header. The repair improves discoverability, while the assertions and real service boundaries remain the actual depth evidence.

Later audits corrected the audit itself#

Wave 5 demonstrated why source review needed a live check: it rejected false assumptions about messages-center refresh data and the Nisaba SSR fallback, corrected unresolved-versus-rejected consent language, removed vacuous notebook guards, and added failure-mode coverage.
Wave 6 found more stale “route absent” claims and a Veritas/Nisaba test aimed at dead UI. The current bridge spec explicitly targets the mounted VeritasSourceDirectory, opens a source profile, follows the real Nisaba lineage link, and verifies the destination surface.
Wave 6 also shipped focused real-HTTP specs for Arete habits, Metis ingest, Sophia answers, Psyche memory, Nisaba notebooks, entitlements, Telegram binding, tenant invites, persona rehearsal, shared-library resolution, and admin auth policy. Those successors—not Wave 4's deferred list—record the delivered expansion.

Boundaries and gaps#

A dated 52-journey audit is not the 64-journey inventory. New journeys, specs, and grade changes must be read from current coverage.md.
A repaired assertion is not an executed assertion today. This page was reconciled from source. It does not claim a new Oshun web E2E run or a current release verdict.
A harness ledger is not production provenance persistence. It proves the inspector's clean, broken, unavailable, and empty contracts without silently depending on a developer database.
Real service-worker activation is not a device matrix. The browser and unit proofs close the false-positive cache condition, but storage eviction, update timing, and installed-shell behavior still vary by browser and OS.
A Covers: header is not coverage. It repairs the traceability triangle; the linked test still has to drive the claimed step and failure mode.
Current deep grades are not retroactive Wave 4 facts. Install, update, first-time, Tara, and multi-device gained later evidence. Offline first use, Veritas evidence trail, and Veritas-to-Nisaba remain partial in the current matrix.
Feature backlog remains product work. Inert review actions, provider or device boundaries, and absent durable workflows should not be converted into passing tests until the product behavior exists.

Re-run evidence#

The repaired Wave 4 seams can be checked serially without implying that every current journey was walked:

bash

PW_BROWSER_CHANNEL=chrome pnpm exec playwright test \
  -c apps/oshun/web/playwright.config.ts \
  apps/oshun/web/e2e/operator-editorial-review-approval.spec.ts \
  apps/oshun/web/e2e/pwa-lifecycle-deepening.spec.ts \
  apps/oshun/web/e2e/citation-drift-resolution.spec.ts \
  apps/oshun/web/e2e/veritas-nisaba-bridge.spec.ts \
  --workers=1

Then run the focused service-worker policy suite and compare the registry rather than copying the historical grade table:

bash

pnpm exec vitest run apps/oshun/web/src/__tests__/pwa-service-worker-policy.test.ts
rg -n 'Backs the journeys? above' apps/oshun/web/e2e -g '*.spec.ts'

A green focused run proves only these repaired seams. A current full-suite verdict requires the repository's complete web E2E gate against a fresh app and BFF stack.

Source trail#

Cross-references#

Open questions#

Which dated full-suite run should become the current 64-journey runtime baseline, and where will its app/BFF revisions and artifacts be retained?
Which registry grade changes require an explicit evidence-delta note so a later reader can distinguish re-earned depth from silent label drift?
Can the provenance fixture contract be complemented by a disposable durable store test without making local E2E depend on developer-owned data?
Which platform/device matrix owns PWA activation, cache eviction, and installed-shell behavior beyond deterministic Chromium coverage?
Should the remaining build-feasible product backlog live in one maintained release gate instead of being copied across historical wave reports?